Zero hands rose when Tony Pierce asked IT professionals whether their organizations got full value from the technology they already owned. The path to artificial intelligence (AI) starts by spending less, he said.
Pierce, field chief technology officer at Splunk, delivered that message at the company's .conf26 user conference in Denver in September. He spoke alongside John Mymryk, account chief technology officer at Microsoft, and Jason Conger, partner chief technology officer at Splunk. Their session, "Become a Frontier Firm: AI, Digital Resilience, and Splunk on Microsoft Azure," focused on how companies can lay the data and operational foundation needed to support AI transformation. ("Frontier Firm" is Microsoft's term for organizations powered by human-agent teams.)
As spending on agentic AI skyrockets, these experts say the organizations that stand to benefit most are not the ones moving fastest. They are the ones that slow down, simplify their environments and use what they already have before buying more.
The gap between AI ambition and readiness is widening. According to Gartner, worldwide spending on AI will total $2.59 trillion in 2026, a 47% increase from the previous year. Yet only 17% of organizations have deployed AI agents, even as more than 60% plan to within two years. Gartner also predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025.
The panelists outlined five considerations for organizations building an agentic AI foundation.
Why Should Organizations Slow Down on Agentic AI?
An agentic AI foundation is the combination of clean data, streamlined operations and governed access controls that organizations need before deploying autonomous AI agents at scale.
For the past several years, much of the conversation around AI has centered on the need for organizations to move as quickly as possible to avoid being left behind. Mymryk said it is more important to build an AI foundation well than to build it quickly.
"You have to go slow to go fast," Mymryk said.
"You want to make sure that your data is good, your operations are great and you build that center of excellence. And you go slow to go fast, so that as you start bringing on the use cases, generating more agents...you have a pretty good grasp of what you're doing."
This does not mean every use case must be perfect before it goes into production. Some organizations may want to embrace the agility of a "fail fast" mindset, Mymryk said. But he said organizations should do so at a pace that allows them to internalize lessons about what works and what does not.
"Start narrow enough to learn," he said. "And then you can optimize and grow."
How Can Organizations Cut Costs on AI?
Pierce pushed back on the idea that AI foundations require heavy investment in all-new technology. Calling himself the "Dave Ramsey of data," after the debt-averse personal finance guru, Pierce said organizations should first look to the resources they already have.
"Don't buy new," Pierce said. "Use what you've got first. Money is a limited resource."
"If you don't need it, don't buy it," he said. "Take a look at what you have."
The benefits of repurposing existing infrastructure go beyond cost savings, Pierce said. New tools carry their own operational burdens, requiring employees to monitor, maintain and learn them.
"Simplify your environment," he said. "That'll also make your AI work better as you deploy your AI."
Does All Data Need to Live in One Place?
It has become conventional wisdom that most organizations do not have an AI problem but rather a data problem. AI tools can ingest and make sense of enormous amounts of data, but that information is often disconnected, inconsistent or otherwise flawed.
Pierce, Mymryk and Conger did not suggest organizations should tolerate data silos. But they said organizations do not necessarily need to unify data in one central repository.
"That got expensive, if you sent everything and anything into the fast, efficient, human-readable index, especially in this AI world," Conger said. "So now we're saying, have it your way. Send it where you need to drive the right action."
Conger pointed to federated search as one way to accomplish this. If data already resides in Azure, for example, organizations can access and correlate it without moving it into Splunk.
"You don't necessarily have to ship data across," Conger said.
How Much Access Should AI Agents Have?
"There are some people in this world that say, 'Well, AI is better the more data you give it,'" Pierce told The Forecast in a post-panel interview. "I'm not that guy."
"You build AI agents to answer a question, to enhance the human," Pierce said. "We don't want to say, 'Here's everything.' Modern ecosystems are so complicated. How are you going to monitor [the environment]? You need to limit it."
Pierce said organizations should not exempt AI agents from the Zero Trust principles they have spent years adopting.
"For the last 10 years, we've been talking about Zero Trust," he said.
"Why did we put a superhuman on the network...and give them permissions through the entire network? We should not do that."
When Is AI Not the Right Answer?
With so much hype surrounding generative AI and now agentic AI over the past several years, it is understandable that organizations are focusing their attention and resources on the technology. But Pierce said other ways to automate workflows may be a better fit for some use cases, including automation agents that follow predefined workflows rather than using their own judgment to complete a task.
He said organizations can use a common visibility layer to aggregate and normalize data, whether they rely on AI, machine learning or human investigators.
"Make the decision: Do I need an AI agent, or do I need an automation agent?" Pierce said. "And know the difference between the two."
Related stories:
Calvin Hennick is a contributing writer. His work appears in BizTech, Engineering Inc., The Boston Globe Magazine and elsewhere. He is also the author of Once More to the Rodeo: A Memoir. Find him on LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
